T08 · Insecure Dependencies
- Location
README.md:17- Finding
Unpinned Installation Command References a Different Skill Package
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This appears to be a coaching skill, but its identity, install instructions, and memory behavior are inconsistent enough that users should review it before installing.
Install only after the publisher aligns the README, manifest, references, repository, and install command with the actual package; fixes version consistency; removes or renames the INTP material; and documents opt-in memory with retention and deletion controls. I found no embedded code execution, credential theft, destructive behavior, or exfiltration instructions.
README.md:17Unpinned Installation Command References a Different Skill Package
The README documents an INTP coaching skill while the manifest says the skill is for ESTJ users. This mismatch can cause users, reviewers, or downstream systems to misunderstand the skill’s actual purpose and behavior, creating a supply-chain style trust issue where the published metadata and documentation do not describe the same artifact.
The documented behavior, examples, and personality guidance are all INTP-specific even though the skill metadata advertises ESTJ coaching. In a coaching context, this is especially risky because users may receive advice optimized for the wrong personality profile, undermining safety, effectiveness, and informed consent about what the skill actually does.
The reference content materially contradicts the ESTJ skill’s stated purpose by describing traits and strengths that align with a different personality profile. In a coaching skill, this can systematically misguide outputs, causing the agent to give inappropriate advice to users seeking personality-specific support and undermining trust and safety in guidance.
These sections repeatedly attribute ESTJ behavior to contradictory functions such as Ti-Ne, Fe weakness, and other patterns inconsistent with the skill’s declared domain. Because this is a behavioral coaching skill, the contradiction is operationally dangerous: the model may internalize and apply the wrong framework, leading to flawed self-development advice, mislabeling of user struggles, and potentially harmful interpersonal recommendations.
The reference file is for INTP coaching while the skill manifest declares an ESTJ coaching purpose, creating a clear scope and behavior mismatch. In an agent skill, this can cause the model to deliver guidance for the wrong personality type, leading to systematically inappropriate advice, reduced reliability, and potential user harm in sensitive self-help or leadership contexts.
The README states that growth records and user profiles are saved automatically, but it does not explain consent, storage location, retention period, deletion options, or privacy handling. Because this skill processes potentially sensitive self-reflection or mental-health-adjacent content, silent persistence increases privacy risk and may violate user expectations or platform policy.
The skill advertises saving user conversations and actions for future sessions without a clear notice about retention, consent, access controls, or deletion. In a coaching context, users are likely to disclose sensitive emotional, behavioral, and professional information, so undisclosed persistence creates privacy and secondary-use risk.
The skill instructs the agent to record core problems, interests, action records, and key insights into a user profile without explaining handling safeguards. This creates a structured dossier of potentially sensitive personal data that could be exposed, misused across sessions, or retained longer than the user expects.
Persisting detailed user disclosures and replaying them in future interactions increases the chance of privacy leakage, especially if the agent later surfaces prior sensitive statements in the wrong context or to the wrong user/session. Natural-language memory is especially risky because stored summaries may contain emotional state, workplace issues, and behavioral patterns that users did not expect to be retained.
The end-of-conversation archiving flow explicitly directs the agent to save user-specific problems, insights, plans, and execution results for future use without a visible privacy warning or consent checkpoint. Because the skill is framed as a supportive coaching tool, users may share candid personal and workplace details, making silent archival more dangerous in context.
The manifest advertises an ESTJ coaching skill, but the keywords and tags describe an INTP-focused coaching tool. This metadata divergence can mislead users, indexing systems, or downstream agents into invoking the wrong skill, causing inappropriate guidance and reducing trust in the skill supply chain. In security-sensitive ecosystems, mislabeled metadata also raises suspicion of repackaging or deceptive distribution.
The metadata indicates the document belongs to a different skill path, suggesting copy-paste reuse or cross-skill contamination. This increases the risk that the wrong reference material is loaded, maintained, or trusted during skill execution, which can propagate incorrect advice and make future maintenance errors more likely.
This markdown file contains all user-facing guidance in Chinese, and nowhere indicates that the skill is intentionally region-specific or that users may opt into another language. Under the policy, forcing a specific language without user choice is a natural-language policy concern.
The boundary section says '不替用户做决定', implying the coach should avoid deciding on behalf of the user. Elsewhere, advisor-mode examples explicitly recommend a preferred option such as '先试方案 1', which creates tension with that stated intent by steering toward a concrete choice.
The manifest description scopes the skill to growth coaching for ESTJ users and ESTJ-specific blockers. In SKILL.md, the '用户不是 ESTJ' section says other personality types can also use the skill, which expands the intended audience beyond the manifest's stated scope.
The repository URL points to a project whose name appears unrelated to ESTJ coaching, which creates provenance ambiguity. Users or automated systems may rely on the repository for source verification, updates, or trust decisions, and a mismatched link can hide the true origin of the skill or frustrate auditing. In this context the issue is suspicious, though the file alone does not prove malicious behavior.
This markdown file is entirely written in Chinese and presents the skill's internal reference material in a single language, with no indication that language selection is optional or user-driven. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation.